1 00:00:02,440 --> 00:00:06,600 Speaker 1: All media, soul of an angel, body of a devil, 2 00:00:06,680 --> 00:00:09,280 Speaker 1: chosen by God and perfected by science. This is better 3 00:00:09,320 --> 00:00:23,560 Speaker 1: offline and I'm your host ed Zitron. Now we're working 4 00:00:23,560 --> 00:00:25,479 Speaker 1: on my newsletter. Last week I was chatting with my 5 00:00:25,520 --> 00:00:28,640 Speaker 1: friend friend of the show casey Kagawa about generative AI, 6 00:00:28,720 --> 00:00:32,000 Speaker 1: and we kept coming back to one thought, where's the money? 7 00:00:32,120 --> 00:00:34,680 Speaker 1: Where is it? Not really where is the money? Where 8 00:00:34,760 --> 00:00:38,640 Speaker 1: is the money that this supposedly revolutionary, world changing industry 9 00:00:38,680 --> 00:00:41,400 Speaker 1: is making and of course we'll make in the future. 10 00:00:42,240 --> 00:00:45,920 Speaker 1: And the answer is simple. After hours of hours of 11 00:00:45,920 --> 00:00:49,000 Speaker 1: grinding through earnings, of grinding through media articles, of grinding 12 00:00:49,000 --> 00:00:51,519 Speaker 1: through all sorts of things, I just don't believe it 13 00:00:51,560 --> 00:00:55,680 Speaker 1: really exists. It's real, but it's small. Generative AI lacks 14 00:00:55,720 --> 00:00:59,240 Speaker 1: the basic unit economics, product market fit, or market penetration 15 00:00:59,320 --> 00:01:02,880 Speaker 1: associated with any meaningful software boom, and outside of open AI, 16 00:01:02,960 --> 00:01:06,360 Speaker 1: the industry may be pathetically hopelessly small or while providing 17 00:01:06,400 --> 00:01:10,200 Speaker 1: few meaningful business returns and constantly losing money. I'm going 18 00:01:10,240 --> 00:01:12,520 Speaker 1: to be pretty straightforward with everything I say in this 19 00:01:12,600 --> 00:01:15,040 Speaker 1: two parter because the numbers and the facts in my 20 00:01:15,120 --> 00:01:18,000 Speaker 1: hypotheses are pretty fucking damning of both the generative AI 21 00:01:18,120 --> 00:01:22,080 Speaker 1: industry and its associated boosters. You're going to get this episode. 22 00:01:22,080 --> 00:01:23,920 Speaker 1: Then there's going to be a monologue about something else 23 00:01:24,040 --> 00:01:26,000 Speaker 1: or something related. I really haven't got to it yet, 24 00:01:26,240 --> 00:01:28,640 Speaker 1: and then a second part which I'm recording immediately after 25 00:01:28,640 --> 00:01:32,120 Speaker 1: this one, a little behind the curtain there for you. Anyway, 26 00:01:32,600 --> 00:01:35,360 Speaker 1: in reporting this analysis, I've done everything I can to 27 00:01:35,400 --> 00:01:37,200 Speaker 1: try and push back against my own work, and I've 28 00:01:37,200 --> 00:01:39,840 Speaker 1: saw evidence to counter the things that I've seen, like 29 00:01:39,880 --> 00:01:43,080 Speaker 1: the revenue and the business models of these companies. Yet 30 00:01:43,200 --> 00:01:45,920 Speaker 1: in doing so, I've only become more convinced of the 31 00:01:46,000 --> 00:01:49,320 Speaker 1: flimsiness of generative AI and the associated industry, and the 32 00:01:49,400 --> 00:01:51,640 Speaker 1: likelihood of this bubble bursting in a way that kneecaps 33 00:01:51,640 --> 00:01:54,920 Speaker 1: tech valuations for a prolonged period, or worse, hits the 34 00:01:54,960 --> 00:01:58,720 Speaker 1: major stock market. Now, I really had originally written a 35 00:01:58,800 --> 00:02:02,040 Speaker 1: far more jocular and perc script, but while I was 36 00:02:02,040 --> 00:02:04,640 Speaker 1: writing it, I realized I really had to be blunt 37 00:02:04,640 --> 00:02:08,640 Speaker 1: because what I'm describing is a systemic failure. Venture capital 38 00:02:08,680 --> 00:02:11,600 Speaker 1: has propped up Open AI and Anthropic, two companies that 39 00:02:11,680 --> 00:02:14,520 Speaker 1: burned are combined ten point five billion dollars in twenty 40 00:02:14,560 --> 00:02:16,960 Speaker 1: twenty four, and that number is set to double or 41 00:02:17,040 --> 00:02:20,120 Speaker 1: more in twenty twenty five. The tech media has allowed 42 00:02:20,160 --> 00:02:23,520 Speaker 1: Sam Mortman to twist them to validate completely fictional ideas 43 00:02:23,880 --> 00:02:27,360 Speaker 1: as a means of propping up this unprofitable, environmentally destructive 44 00:02:27,360 --> 00:02:30,600 Speaker 1: software company, and big tech has become so disconnected from 45 00:02:30,680 --> 00:02:33,840 Speaker 1: reality that it is incapable of seeing how little actual 46 00:02:33,880 --> 00:02:37,240 Speaker 1: returns there are in generative AI, and they're failing. By 47 00:02:37,280 --> 00:02:40,200 Speaker 1: the way. As I'll walk you through in these episodes, 48 00:02:40,240 --> 00:02:43,840 Speaker 1: the GENERATIVEAI industry is very small, with the consumer market 49 00:02:43,840 --> 00:02:47,280 Speaker 1: of the entire American GENERATIVEAI industry outside of chat GBT 50 00:02:47,680 --> 00:02:51,520 Speaker 1: barely cracking one hundred million monthly active users, which puts 51 00:02:51,520 --> 00:02:53,880 Speaker 1: them below a lot of free to play games that 52 00:02:53,919 --> 00:02:57,760 Speaker 1: you get on your iPhone. Hyperscalers have already spent hundreds 53 00:02:57,800 --> 00:03:00,359 Speaker 1: of billions of dollars in capital expenditures for an AI 54 00:03:00,400 --> 00:03:03,280 Speaker 1: industry that has the combined monthly active users of a 55 00:03:03,360 --> 00:03:06,399 Speaker 1: free to play mobile game. I really must repeat myself, 56 00:03:06,600 --> 00:03:11,800 Speaker 1: it's insane, But unlike most mobile games, GENERATIVEAI doesn't really 57 00:03:11,840 --> 00:03:14,960 Speaker 1: make any money. And for those of you wondering if 58 00:03:15,000 --> 00:03:18,040 Speaker 1: selling access to AI models is the solution, it's important 59 00:03:18,040 --> 00:03:20,760 Speaker 1: to know that open AI the market leader in generative AI, 60 00:03:21,040 --> 00:03:23,280 Speaker 1: made less than a billion dollars an API calls in 61 00:03:23,320 --> 00:03:25,959 Speaker 1: twenty twenty four, and that's when people plug their models 62 00:03:26,000 --> 00:03:27,760 Speaker 1: them For those of you who don't understand, so it's 63 00:03:27,800 --> 00:03:30,320 Speaker 1: the difference between you to load up the chat GPT 64 00:03:30,440 --> 00:03:34,160 Speaker 1: app or someone has an AI generative AI like chat 65 00:03:34,160 --> 00:03:37,520 Speaker 1: gpt plugged into it. Now, Microsoft pays open Ai a 66 00:03:37,560 --> 00:03:40,920 Speaker 1: revenue share of twenty percent on them selling open AI's models, 67 00:03:40,960 --> 00:03:44,680 Speaker 1: so two hundred billion dollars. So this means that Microsoft 68 00:03:44,760 --> 00:03:46,560 Speaker 1: likely when he makes a billion dollars in revenue from 69 00:03:46,600 --> 00:03:50,160 Speaker 1: API calls themselves, this is a pathetic amount of money 70 00:03:50,160 --> 00:03:52,800 Speaker 1: and suggests there really isn't significant demand at all or 71 00:03:53,520 --> 00:03:57,080 Speaker 1: they're not charging enough. Neither of these are great, And honestly, 72 00:03:57,080 --> 00:03:59,120 Speaker 1: I'm sick and tired of hearing people prop up this 73 00:03:59,200 --> 00:04:02,600 Speaker 1: fucking industry. In these episodes, I will explain as calmly 74 00:04:02,640 --> 00:04:05,760 Speaker 1: as possible how the generative AI industry barely exists outside 75 00:04:05,800 --> 00:04:09,000 Speaker 1: of open Ai, And honestly, in writing this, I've become 76 00:04:09,000 --> 00:04:12,480 Speaker 1: completely disgusted at Silicon Valley at the waste. Why is 77 00:04:12,560 --> 00:04:17,200 Speaker 1: nobody talking about the revenues, Why is nobody sharing real 78 00:04:17,279 --> 00:04:20,400 Speaker 1: user numbers other than open Ai. Well, I believe it's 79 00:04:20,440 --> 00:04:22,960 Speaker 1: because there isn't that much money, and there certainly aren't 80 00:04:22,960 --> 00:04:26,839 Speaker 1: that many users. Nobody is making a profit from this 81 00:04:26,920 --> 00:04:29,920 Speaker 1: other than consultants, and that's because this is a hype 82 00:04:30,080 --> 00:04:34,200 Speaker 1: driven movement. What you see on TV and in the 83 00:04:34,240 --> 00:04:37,480 Speaker 1: newspaper is not the advent of a revolutionary piece of technology. 84 00:04:37,760 --> 00:04:41,160 Speaker 1: It's a cynical marketing campaign for one company, open Ai. 85 00:04:42,560 --> 00:04:45,440 Speaker 1: I need you to understand how precarious this all is. 86 00:04:46,160 --> 00:04:48,640 Speaker 1: So much money has been wasted propping up an industry 87 00:04:48,640 --> 00:04:51,799 Speaker 1: that only burns money, that does not have mass market appeal. 88 00:04:52,360 --> 00:04:55,839 Speaker 1: Chat GPT is not significant enough, or useful enough, or 89 00:04:55,880 --> 00:04:59,880 Speaker 1: meaningful enough to justify spending nine billion dollars to lose 90 00:05:00,000 --> 00:05:02,960 Speaker 1: five billion dollars. And yes, those are the raw economics 91 00:05:02,960 --> 00:05:05,960 Speaker 1: of open Ai. Now you may say, well, Ubber lost 92 00:05:05,960 --> 00:05:08,600 Speaker 1: a lot of money, didn't it, Edward? Guess what? Uber 93 00:05:08,640 --> 00:05:10,720 Speaker 1: lost about six point two billion dollars in one year, 94 00:05:10,760 --> 00:05:12,839 Speaker 1: and that was in twenty twenty when they couldn't run 95 00:05:12,839 --> 00:05:16,200 Speaker 1: their bloody service. Uber is a very different company. I 96 00:05:16,279 --> 00:05:18,279 Speaker 1: will gladly if you email me, I will explain this 97 00:05:18,320 --> 00:05:20,359 Speaker 1: to you. But Uber is not a comparison. There is 98 00:05:20,400 --> 00:05:23,240 Speaker 1: no comparison to what open ai is doing, what Anthropic 99 00:05:23,320 --> 00:05:26,080 Speaker 1: is doing. I sound crazy as ever, but you're going 100 00:05:26,160 --> 00:05:29,760 Speaker 1: to understand why when I'm done. I am deeply worried 101 00:05:29,760 --> 00:05:32,000 Speaker 1: about this industry, and I need you to know why. 102 00:05:33,080 --> 00:05:36,039 Speaker 1: But in this first episode, I'm going to focus on 103 00:05:36,040 --> 00:05:38,720 Speaker 1: one specific thing, and that's the capitalist delusion known as 104 00:05:38,760 --> 00:05:42,720 Speaker 1: open Ai, a company that encompasses almost all of the traffic, funding, 105 00:05:42,760 --> 00:05:46,240 Speaker 1: and attention in generative AI, and I believe they die 106 00:05:46,279 --> 00:05:50,080 Speaker 1: without a constant flow of venture capital and hyperscale or welfare. 107 00:05:50,120 --> 00:05:52,440 Speaker 1: And I actually don't know why I said I believe 108 00:05:52,480 --> 00:05:57,440 Speaker 1: that they will. They cannot survive without that money. But okay, 109 00:05:57,560 --> 00:06:01,200 Speaker 1: let's take a second. Let's talk about open as unit economics. 110 00:06:02,000 --> 00:06:04,520 Speaker 1: Putting aside the high the blaster, open Ai, as with 111 00:06:04,560 --> 00:06:08,640 Speaker 1: all GENERATIVAI model developers, loses money on every single prompt 112 00:06:08,680 --> 00:06:12,159 Speaker 1: and output. Its products do not scale like traditional software, 113 00:06:12,240 --> 00:06:14,680 Speaker 1: in that the more users it gets, the more expensive 114 00:06:14,760 --> 00:06:17,039 Speaker 1: its services are to run because its models are so 115 00:06:17,120 --> 00:06:21,760 Speaker 1: compute intensive. For example, Chat GPT having four hundred million 116 00:06:21,800 --> 00:06:23,880 Speaker 1: weekly active users is not the same thing as a 117 00:06:23,880 --> 00:06:26,839 Speaker 1: traditional app like Instagram or Facebook having that many users. 118 00:06:26,920 --> 00:06:29,960 Speaker 1: Or indeed Uber. The cost of serving a regular user 119 00:06:29,960 --> 00:06:32,800 Speaker 1: of an app like Instagram is significantly smaller because these 120 00:06:32,880 --> 00:06:37,720 Speaker 1: are effectively websites with connecting APIs, images, videos, and user interactions. 121 00:06:38,000 --> 00:06:41,120 Speaker 1: These platforms aren't innately compute heavy, and so you don't 122 00:06:41,160 --> 00:06:44,520 Speaker 1: need to have the same level of infrastructure to support 123 00:06:44,560 --> 00:06:48,440 Speaker 1: the same amount of people. Conversely, GENERATIVAI requires expensive to 124 00:06:48,440 --> 00:06:51,560 Speaker 1: buy and expensive to run and expensive to maintain graphics 125 00:06:51,600 --> 00:06:55,960 Speaker 1: processing units GPUs, both for inference and training the models themselves. 126 00:06:56,480 --> 00:06:58,720 Speaker 1: These GPUs must be run at full tilt for both 127 00:06:58,720 --> 00:07:02,039 Speaker 1: inference and training, which organs their lifespan while also consuming 128 00:07:02,120 --> 00:07:05,000 Speaker 1: ungodly amounts of energy. And by the way, inference is 129 00:07:05,080 --> 00:07:07,440 Speaker 1: just the thing that happens when you tell chat GPTs. 130 00:07:07,480 --> 00:07:09,200 Speaker 1: I think it infers the meaning of the prompt. And 131 00:07:09,640 --> 00:07:11,640 Speaker 1: the training is what they do when they throw all 132 00:07:11,640 --> 00:07:14,560 Speaker 1: the training data to make the model huh smart. Not 133 00:07:14,600 --> 00:07:17,520 Speaker 1: really though, And by the way, surrounding the GPU in 134 00:07:17,560 --> 00:07:19,320 Speaker 1: there isn't that the GPUs just kind of hang out. 135 00:07:19,320 --> 00:07:21,760 Speaker 1: There's the rest of a computer, which is usually highly 136 00:07:21,760 --> 00:07:25,600 Speaker 1: specked and incredibly difficult to cool and thus very very expensive. 137 00:07:26,480 --> 00:07:30,000 Speaker 1: These generative AI models also require endless amounts of training 138 00:07:30,080 --> 00:07:32,600 Speaker 1: data and supplies of that training data have been running 139 00:07:32,600 --> 00:07:35,640 Speaker 1: out for a long time. While synthetic data might bridge 140 00:07:35,640 --> 00:07:38,040 Speaker 1: some of the gap, there are likely diminishing returns due 141 00:07:38,040 --> 00:07:40,080 Speaker 1: to the sheer amount of data necessary to make a large, 142 00:07:40,160 --> 00:07:43,200 Speaker 1: loud and quich model even larger, as much as more 143 00:07:43,240 --> 00:07:47,920 Speaker 1: than four times the size of the Internet. This is insane. 144 00:07:48,120 --> 00:07:50,960 Speaker 1: There is not enough data, and it already kind of sucks, 145 00:07:51,120 --> 00:07:55,120 Speaker 1: and it's not getting better. These companies also must spend 146 00:07:55,240 --> 00:07:57,760 Speaker 1: hundreds of millions of dollars on salaries to attract and 147 00:07:57,800 --> 00:08:00,880 Speaker 1: retain AI talent, as much as one point five billion 148 00:08:00,920 --> 00:08:03,119 Speaker 1: dollars a year in open AI's case, and that's before 149 00:08:03,160 --> 00:08:07,120 Speaker 1: stock based compensation. In twenty sixteen, Microsoft claimed that top 150 00:08:07,160 --> 00:08:09,720 Speaker 1: AI talent could cost as much as an NFL quarterback 151 00:08:09,720 --> 00:08:12,360 Speaker 1: to hire, and that some has likely only increased since 152 00:08:12,400 --> 00:08:15,280 Speaker 1: then given the GENERATIVAI frenzy and the fact they're overpaying 153 00:08:15,360 --> 00:08:18,360 Speaker 1: quarterbacks as in the side. One analyst told The Wall 154 00:08:18,360 --> 00:08:21,000 Speaker 1: Street Journal that companies running generative AI models could and 155 00:08:21,080 --> 00:08:24,560 Speaker 1: I quote be utilizing half their capital expenditures because all 156 00:08:24,600 --> 00:08:29,280 Speaker 1: of these things could break down. That's that's really bad. 157 00:08:29,400 --> 00:08:32,079 Speaker 1: These costs are not a burden on open ai or Anthropic, 158 00:08:32,120 --> 00:08:35,520 Speaker 1: but they absolutely are on Microsoft, Google and Amazon. This 159 00:08:35,600 --> 00:08:39,000 Speaker 1: shit's crazy anyway. As a result of the costs of 160 00:08:39,080 --> 00:08:41,520 Speaker 1: running these services, a free user of Chat GPT is 161 00:08:41,559 --> 00:08:44,680 Speaker 1: a cost burden on open Ai, as is every single 162 00:08:44,720 --> 00:08:48,520 Speaker 1: free customer of Google's Gemini, Anthropics, Clawed, Perplexity, or any 163 00:08:48,520 --> 00:08:52,200 Speaker 1: other generative AI company. Said costs are also so severe 164 00:08:52,200 --> 00:08:56,440 Speaker 1: that even paying customers lose these companies' money. Even the 165 00:08:56,440 --> 00:08:58,680 Speaker 1: most successful company in the business appears to have no 166 00:08:58,760 --> 00:09:01,520 Speaker 1: way to stop burning cap And as I'll explain, there's 167 00:09:01,559 --> 00:09:05,560 Speaker 1: only really one real company in the industry, Open Ai, 168 00:09:06,160 --> 00:09:09,640 Speaker 1: and open ai is not a real business. But let's 169 00:09:09,720 --> 00:09:12,559 Speaker 1: start with a really, really important fact. If you forget 170 00:09:12,600 --> 00:09:15,600 Speaker 1: everything I say, I want you to remember this. Open 171 00:09:15,640 --> 00:09:19,160 Speaker 1: Ai spent nine billion dollars to make just under four 172 00:09:19,200 --> 00:09:22,079 Speaker 1: billion dollars in twenty twenty four, and the entirety of 173 00:09:22,120 --> 00:09:25,720 Speaker 1: their revenue that's about four billion dollars, is spent on compute, 174 00:09:25,840 --> 00:09:28,360 Speaker 1: two billion dollars to run models and three billion dollars 175 00:09:28,440 --> 00:09:31,680 Speaker 1: to train them. That is completely and utterly fucking insane. 176 00:09:31,880 --> 00:09:35,240 Speaker 1: That is bonkers. That is crazy. That is completely nuts. 177 00:09:35,400 --> 00:09:37,680 Speaker 1: This is not a real company. It is insane. We're 178 00:09:37,720 --> 00:09:40,880 Speaker 1: allowing this. Everyone should be screaming this at everyone. We 179 00:09:40,960 --> 00:09:43,719 Speaker 1: live in an alternate reality where this is acceptable. There 180 00:09:43,760 --> 00:09:46,320 Speaker 1: has been no precedent for this. Not Amazon Web Services, 181 00:09:46,360 --> 00:09:49,240 Speaker 1: not Uber, not anyone. No one has done this, and 182 00:09:49,280 --> 00:09:52,000 Speaker 1: it's sickening and wasteful that we continue to. And in 183 00:09:52,040 --> 00:09:54,600 Speaker 1: the past I've repeatedly said that open ai lost five 184 00:09:54,600 --> 00:09:57,440 Speaker 1: billion dollars after revenue. Now that is true. By the way, 185 00:09:57,720 --> 00:10:00,320 Speaker 1: that is completely true. They made money and they lost money, 186 00:10:00,320 --> 00:10:03,920 Speaker 1: but ended up losing five billion either either way. However, 187 00:10:04,000 --> 00:10:06,640 Speaker 1: I really just I can't in good conscience suggest that 188 00:10:06,679 --> 00:10:10,080 Speaker 1: open ai only spent five billion dollars. It's time to 189 00:10:10,080 --> 00:10:12,319 Speaker 1: be honest about these numbers. While it's fair to say 190 00:10:12,360 --> 00:10:15,880 Speaker 1: that their net losses are five billion, they spent nine 191 00:10:15,920 --> 00:10:30,880 Speaker 1: billion dollars to lose five billion dollars. Let's really get 192 00:10:30,880 --> 00:10:33,320 Speaker 1: down into the nitty gritty of these numbers. So, as 193 00:10:33,360 --> 00:10:36,200 Speaker 1: discussed previously, according to the reporting by the Information, open 194 00:10:36,240 --> 00:10:38,320 Speaker 1: AI's revenue was likely somewhere in the region of four 195 00:10:38,320 --> 00:10:41,840 Speaker 1: billion dollars in twenty twenty four. Their burn ray, according 196 00:10:41,880 --> 00:10:45,000 Speaker 1: to the Information, was five billion dollars after revenue in 197 00:10:45,000 --> 00:10:48,560 Speaker 1: twenty twenty four, excluding stock based compensation, which open Ai, 198 00:10:48,760 --> 00:10:51,200 Speaker 1: like other startups, uses as a means of compensation on 199 00:10:51,240 --> 00:10:54,080 Speaker 1: top of cash. Nevertheless, the more it gives away, the 200 00:10:54,160 --> 00:10:57,439 Speaker 1: less it has for capital raises. And these are technically costs, 201 00:10:57,520 --> 00:11:00,680 Speaker 1: though they're not real money unless there's a liquidity event, 202 00:11:00,720 --> 00:11:02,280 Speaker 1: but that's our you don't need to worry about that. 203 00:11:02,960 --> 00:11:05,280 Speaker 1: To put this in blunt terms, based on reporting by 204 00:11:05,280 --> 00:11:08,319 Speaker 1: the Information, and I'm repeating myself here, but I really 205 00:11:08,360 --> 00:11:11,720 Speaker 1: need you to remember this, running open Ai cost nine 206 00:11:12,080 --> 00:11:15,800 Speaker 1: billion dollars in twenty twenty four. The cost of the computer, 207 00:11:16,200 --> 00:11:19,360 Speaker 1: to the compute to train models alone three billion dollars 208 00:11:19,440 --> 00:11:23,480 Speaker 1: obliterates the entirety of their subscription revenue, which was about 209 00:11:23,480 --> 00:11:25,559 Speaker 1: three billion dollars by the way, and the compute from 210 00:11:25,600 --> 00:11:28,160 Speaker 1: running models two billion dollars takes the rest and then some. 211 00:11:28,600 --> 00:11:31,920 Speaker 1: They actually end up losing an extra billion on top 212 00:11:31,920 --> 00:11:34,040 Speaker 1: of that. Sam Altman's net worth is a billion dollars. 213 00:11:34,080 --> 00:11:36,600 Speaker 1: By the way, Casey Gogawa has now used this as 214 00:11:36,640 --> 00:11:39,560 Speaker 1: the the Altman Index, so it's like you've lost one 215 00:11:39,600 --> 00:11:43,200 Speaker 1: Sam Altman. That's a billion dollars. But just to be clear, 216 00:11:43,240 --> 00:11:45,720 Speaker 1: it doesn't just cost more to run open Ai than 217 00:11:45,720 --> 00:11:47,679 Speaker 1: they make. It costs them a billion dollars more than 218 00:11:47,679 --> 00:11:50,000 Speaker 1: the entirety of their revenue to run the software they 219 00:11:50,040 --> 00:11:54,320 Speaker 1: sell before any other costs. Why are we not more 220 00:11:54,360 --> 00:11:58,240 Speaker 1: concerned about this company now? Something else to note is 221 00:11:58,240 --> 00:12:00,800 Speaker 1: that open ai also spends another line amount of money 222 00:12:00,800 --> 00:12:03,560 Speaker 1: on salaries, over seven hundred million dollars in twenty twenty 223 00:12:03,559 --> 00:12:07,000 Speaker 1: four before you consider that compensation from stock, a number 224 00:12:07,000 --> 00:12:09,839 Speaker 1: that will also have to increase because open ai is growing, 225 00:12:09,880 --> 00:12:12,320 Speaker 1: which means hiring as many people as possible, and they're 226 00:12:12,320 --> 00:12:15,439 Speaker 1: paying through the nose for them. But let's talk about 227 00:12:15,440 --> 00:12:18,760 Speaker 1: how open ai makes money. Open Ai sells access to 228 00:12:18,800 --> 00:12:21,719 Speaker 1: its models via its API and selling premium subscriptions to 229 00:12:21,800 --> 00:12:25,000 Speaker 1: chat gpt. The majority of its revenue over seventy percent, 230 00:12:25,080 --> 00:12:28,560 Speaker 1: comes from subscriptions to premium versions of chat GPT. The 231 00:12:28,640 --> 00:12:31,600 Speaker 1: Information also reported that open ai now has fifteen point 232 00:12:31,679 --> 00:12:34,760 Speaker 1: five million paying customers, though it's unclear what level of 233 00:12:34,800 --> 00:12:37,439 Speaker 1: the service they're paying for or how sticky these customers 234 00:12:37,480 --> 00:12:39,319 Speaker 1: are is in how luckily they are to stick around, 235 00:12:39,440 --> 00:12:41,760 Speaker 1: or the cost of acquiring these customers are really any 236 00:12:41,760 --> 00:12:44,400 Speaker 1: other metric to tell them tell us how valuable these 237 00:12:44,400 --> 00:12:47,960 Speaker 1: customers are to the bottom line. Nevertheless, open ai loses 238 00:12:48,000 --> 00:12:52,200 Speaker 1: money on every single paying customer, just like its free users. 239 00:12:52,760 --> 00:12:56,880 Speaker 1: Increasing paid subscribers to open ai services somehow increases open 240 00:12:56,920 --> 00:13:00,640 Speaker 1: AI's burn rate. This is not a real company. Now, 241 00:13:00,640 --> 00:13:02,800 Speaker 1: The New York Times reports the open ai projects it 242 00:13:02,840 --> 00:13:05,800 Speaker 1: will make eleven point six billion dollars in twenty twenty five, 243 00:13:06,000 --> 00:13:08,120 Speaker 1: and assuming that open ai burns at the same rate 244 00:13:08,200 --> 00:13:10,240 Speaker 1: it did in twenty twenty four, spending two point two 245 00:13:10,360 --> 00:13:12,640 Speaker 1: five dollars to make one dollar, open ai is on 246 00:13:12,760 --> 00:13:14,920 Speaker 1: course to bone over twenty six billion dollars in twenty 247 00:13:14,960 --> 00:13:17,960 Speaker 1: twenty five for a loss of fourteen point four billion dollars. 248 00:13:18,360 --> 00:13:21,439 Speaker 1: Who knows what their actual cost will be. Now you've 249 00:13:21,480 --> 00:13:24,000 Speaker 1: probably heard about soft Bank coming in. Soft Bank's going 250 00:13:24,080 --> 00:13:25,680 Speaker 1: to feed the money, and soft banks said they're going 251 00:13:25,760 --> 00:13:28,160 Speaker 1: to spend money on this, that, and the other. That 252 00:13:28,280 --> 00:13:31,960 Speaker 1: round has not closed yet. Masayoshi Son a complete fucking 253 00:13:32,000 --> 00:13:35,280 Speaker 1: idiot who's lost thirty odd billion dollars for soft Bank, 254 00:13:35,320 --> 00:13:40,080 Speaker 1: the Japanese make a conglomerate. He's dedicating billions of dollars 255 00:13:40,080 --> 00:13:43,360 Speaker 1: of revenue to buying open ai services, unless this is 256 00:13:43,400 --> 00:13:45,640 Speaker 1: a straight up trade where he's just sending money before 257 00:13:45,679 --> 00:13:47,920 Speaker 1: the services come in. I don't know if it happens, 258 00:13:48,320 --> 00:13:50,240 Speaker 1: and I'm going to get into things like agents later, 259 00:13:50,280 --> 00:13:53,360 Speaker 1: but the information reported that open ai expects to make 260 00:13:53,400 --> 00:13:57,200 Speaker 1: three billion dollars in revenue from agents. By the end 261 00:13:57,200 --> 00:13:59,240 Speaker 1: of this episode. You're going to realize how fucking stupid 262 00:13:59,240 --> 00:14:02,719 Speaker 1: that sounds. We'll get their layer. It's also important to 263 00:14:02,800 --> 00:14:05,160 Speaker 1: note that open AI's costs are partially subsidized by its 264 00:14:05,200 --> 00:14:08,200 Speaker 1: relationship with Microsoft, which provides cloud compute credits for its 265 00:14:08,320 --> 00:14:11,439 Speaker 1: zero cloud service. Not super technical, it's just when they 266 00:14:11,440 --> 00:14:15,080 Speaker 1: host people's software and files and such and the compute 267 00:14:15,160 --> 00:14:18,720 Speaker 1: to run these models, and they also offer this a steep, 268 00:14:18,760 --> 00:14:21,640 Speaker 1: steep discount to open Ai. Or put another way, it's 269 00:14:21,680 --> 00:14:23,640 Speaker 1: like open ai got paid with air miles, but the 270 00:14:23,680 --> 00:14:26,000 Speaker 1: airline lowered the redemption cost of booking a flight with 271 00:14:26,000 --> 00:14:28,240 Speaker 1: those air miles, allowing it to take more flights than 272 00:14:28,280 --> 00:14:31,800 Speaker 1: any other person with the equivalent amount of points. Until recently, 273 00:14:31,880 --> 00:14:35,400 Speaker 1: open ai exclusively used Microsoft as Zuo services to train 274 00:14:35,520 --> 00:14:38,120 Speaker 1: host and run its models, but recent changes to its 275 00:14:38,120 --> 00:14:40,240 Speaker 1: deal means that open ai is now working with Oracle 276 00:14:40,240 --> 00:14:42,240 Speaker 1: to build up further data centers to train host and 277 00:14:42,320 --> 00:14:45,680 Speaker 1: run its models. It's unclear whether this partnership will work 278 00:14:45,680 --> 00:14:47,720 Speaker 1: in the same way as the Microsoft deal with open 279 00:14:47,720 --> 00:14:51,280 Speaker 1: Ai provided credits and discounts like before. If not, open 280 00:14:51,320 --> 00:14:54,920 Speaker 1: AI's operating costs will only go up. For previous reporting 281 00:14:54,960 --> 00:14:57,680 Speaker 1: from the information, open Ai pays just over twenty five 282 00:14:57,680 --> 00:15:00,240 Speaker 1: percent of the cost of a zero's GPU compute part 283 00:15:00,280 --> 00:15:02,760 Speaker 1: of their deal with Microsoft, and that's about a dollar 284 00:15:02,840 --> 00:15:05,560 Speaker 1: thirty per GPU per hour versus the regular is your 285 00:15:05,600 --> 00:15:07,840 Speaker 1: cost of three dollars and forty cents to four dollars 286 00:15:07,920 --> 00:15:10,480 Speaker 1: an hour. I know that this sounds really technical, but 287 00:15:11,400 --> 00:15:14,160 Speaker 1: in very short they're getting a sweet deal from Microsoft, 288 00:15:14,200 --> 00:15:16,640 Speaker 1: and if anything happens in that, they're completely fucked. They're 289 00:15:16,640 --> 00:15:19,760 Speaker 1: fucked anyway. They don't have They're burning billions of dollars. 290 00:15:19,800 --> 00:15:23,680 Speaker 1: It's insane. But let's talk about user numbers, because open 291 00:15:23,680 --> 00:15:26,360 Speaker 1: ai has quite a few. They recently announced that they 292 00:15:26,400 --> 00:15:30,320 Speaker 1: have four hundred million weekly active users. Now, weekly active 293 00:15:30,400 --> 00:15:32,720 Speaker 1: users is a wanky number and a very strange one 294 00:15:32,720 --> 00:15:35,520 Speaker 1: for a company like this. Open Ai may pretend to 295 00:15:35,520 --> 00:15:38,120 Speaker 1: be a consumer company, but the majority of their revenue 296 00:15:38,120 --> 00:15:41,000 Speaker 1: comes from monthly subscriptions, making them kind of a cloud 297 00:15:41,040 --> 00:15:45,160 Speaker 1: software company. Classically cloud software companies report monthly active users. 298 00:15:45,440 --> 00:15:48,880 Speaker 1: That way, you can, I don't know, compare one number 299 00:15:48,920 --> 00:15:50,920 Speaker 1: which is the amount of active users you have with 300 00:15:51,000 --> 00:15:52,880 Speaker 1: the paid users you have, and then say, oh, that's 301 00:15:52,880 --> 00:15:55,200 Speaker 1: a good business. That's a good business, right there. Man, 302 00:15:55,960 --> 00:15:58,480 Speaker 1: guess what open ai isn't given their monthly active users. 303 00:15:58,760 --> 00:16:02,560 Speaker 1: Don't worry, I might estimated it. When I asked open 304 00:16:02,600 --> 00:16:04,800 Speaker 1: ai to define what a weekly active user was, it 305 00:16:04,840 --> 00:16:07,120 Speaker 1: responded by pointing me to a tweet by chief operating 306 00:16:07,120 --> 00:16:09,920 Speaker 1: officer Brad light Cap that said chat gpt recently crossed 307 00:16:09,920 --> 00:16:12,920 Speaker 1: four one hundred million weekly active users. We feel very 308 00:16:12,960 --> 00:16:15,200 Speaker 1: fortunate serve five percent of the world every week. What 309 00:16:15,320 --> 00:16:19,840 Speaker 1: a fucking liar. It's extremely questionable that open ai refuses 310 00:16:19,880 --> 00:16:21,880 Speaker 1: to define this core metric. By the way, and without 311 00:16:21,880 --> 00:16:23,800 Speaker 1: a definition, in my opinion, there is no way to 312 00:16:23,840 --> 00:16:27,880 Speaker 1: assume anything other than open ai is actively gaming its numbers. Now. 313 00:16:27,920 --> 00:16:30,800 Speaker 1: There's likely two reasons they focus on weekly active users. 314 00:16:31,000 --> 00:16:33,320 Speaker 1: One has described these numbers are easy to game, you 315 00:16:33,320 --> 00:16:36,360 Speaker 1: can choose any seven day period. And also the majority 316 00:16:36,360 --> 00:16:39,120 Speaker 1: of open AI's revenue comes from paid subscriptions to chat gpt. 317 00:16:39,480 --> 00:16:41,800 Speaker 1: And that latter point is crucial because it suggests open 318 00:16:41,840 --> 00:16:44,000 Speaker 1: ai is not doing anywhere near as well as it 319 00:16:44,040 --> 00:16:47,000 Speaker 1: seems based on the very basic metrics used to measure 320 00:16:47,000 --> 00:16:50,200 Speaker 1: the success of a software product. The information reported on 321 00:16:50,280 --> 00:16:53,080 Speaker 1: January thirty first, open Ai, like I mentioned, had fifteen 322 00:16:53,080 --> 00:16:56,840 Speaker 1: point five million monthly paying subscribers, and they added in 323 00:16:56,880 --> 00:16:59,480 Speaker 1: this piece that this was less than a five percent 324 00:16:59,520 --> 00:17:03,480 Speaker 1: conversion rate of open aiy's weekly active users, a statement 325 00:17:03,520 --> 00:17:05,880 Speaker 1: that's kind of like dividing the number fifty two budd 326 00:17:05,880 --> 00:17:08,720 Speaker 1: a letter A. This is not an honest or reasonable 327 00:17:08,760 --> 00:17:11,959 Speaker 1: way to evaluate the success of chat GPT's still unprofitable 328 00:17:11,960 --> 00:17:15,520 Speaker 1: software business, because the actual metric, like I mentioned, would 329 00:17:15,520 --> 00:17:19,200 Speaker 1: have been to definde paying subscribers by monthly active users 330 00:17:19,280 --> 00:17:21,000 Speaker 1: or the other way around. I guess a number that 331 00:17:21,040 --> 00:17:24,199 Speaker 1: would be considerably higher than four hundred million. And the 332 00:17:24,240 --> 00:17:25,720 Speaker 1: reason they don't need to do that, by the way, 333 00:17:25,760 --> 00:17:27,840 Speaker 1: is because you would divide them and see that they 334 00:17:27,840 --> 00:17:30,119 Speaker 1: have a piss poor conversion rate. Good conversion rate is 335 00:17:30,359 --> 00:17:32,920 Speaker 1: way higher than five percent, by the way, and theirs 336 00:17:33,000 --> 00:17:35,600 Speaker 1: is definitely lower. But don't worry. I'm a sneaky little shit. 337 00:17:35,680 --> 00:17:37,119 Speaker 1: So I went and looked some stuff up, and I 338 00:17:37,200 --> 00:17:39,840 Speaker 1: talked to some people based on data from the market 339 00:17:39,840 --> 00:17:43,000 Speaker 1: intelligence firm Center Tower. Open AI's chat gpt app on 340 00:17:43,040 --> 00:17:45,440 Speaker 1: Android and iOS is estimated to have more than three 341 00:17:45,560 --> 00:17:48,600 Speaker 1: hundred and thirty nine million monthly active users, and based 342 00:17:48,600 --> 00:17:52,160 Speaker 1: on traffic data for market intelligence company similar web chatgpt 343 00:17:52,480 --> 00:17:55,000 Speaker 1: dot com had two hundred and forty six million unique 344 00:17:55,000 --> 00:17:57,720 Speaker 1: monthly visitors, and these were in January twenty twenty five. 345 00:17:58,240 --> 00:18:00,679 Speaker 1: There's likely some crossover with people using both the mobile 346 00:18:00,680 --> 00:18:03,359 Speaker 1: and web interfaces, though how big nut group is is 347 00:18:03,440 --> 00:18:06,760 Speaker 1: kind of hard to tell and remains uncertain. Though every 348 00:18:06,840 --> 00:18:09,840 Speaker 1: single person that visits chatgpt dot com might not become 349 00:18:09,880 --> 00:18:12,399 Speaker 1: a user, it's safe to assume that chat GPT's monthly 350 00:18:12,480 --> 00:18:14,879 Speaker 1: active users are somewhere in the region of five hundred 351 00:18:15,119 --> 00:18:18,879 Speaker 1: to six hundred million. That's good, right, Its actual users 352 00:18:18,880 --> 00:18:22,359 Speaker 1: are higher than officially claimed, right, that's good. No, it's bad. 353 00:18:22,720 --> 00:18:25,760 Speaker 1: First of all, each user that uses chat gpt for 354 00:18:25,760 --> 00:18:28,360 Speaker 1: free is a drain on the company, whether they're free 355 00:18:28,440 --> 00:18:30,719 Speaker 1: or not, honestly, but either way, their free ones definitely are. 356 00:18:31,040 --> 00:18:33,320 Speaker 1: It would also suggest that the real conversion rate is 357 00:18:33,359 --> 00:18:36,120 Speaker 1: somewhere in the neighborhood of two point five eight three 358 00:18:36,160 --> 00:18:39,159 Speaker 1: percent from free paid users on chat GPT, which is 359 00:18:39,200 --> 00:18:42,520 Speaker 1: astonishingly bad, and it's a fact that's made worse by 360 00:18:42,560 --> 00:18:44,600 Speaker 1: the fact that every single user, regardless of whether they 361 00:18:44,640 --> 00:18:48,320 Speaker 1: pay or not, loses the money either way. And while 362 00:18:48,359 --> 00:18:50,880 Speaker 1: it's quite common for Silicon Valley companies to play fast 363 00:18:50,880 --> 00:18:54,040 Speaker 1: than loose with metrics, this particularly one is well was 364 00:18:54,080 --> 00:18:56,920 Speaker 1: deeply concerning, and I hypothesize that open ay is choosing 365 00:18:56,960 --> 00:18:59,280 Speaker 1: to go with the weekly versus monthly active users in 366 00:18:59,320 --> 00:19:02,320 Speaker 1: an intentional attempt to avoid people calculating the conversion rate 367 00:19:02,320 --> 00:19:05,720 Speaker 1: of its subscription products. As I will continue to repeat, 368 00:19:05,840 --> 00:19:09,520 Speaker 1: these subscription products lose the company money every single time. 369 00:19:10,280 --> 00:19:12,840 Speaker 1: Now let's talk product strategy, shall we, because I don't 370 00:19:12,840 --> 00:19:16,400 Speaker 1: think open ai really has one. Open Ai makes most 371 00:19:16,400 --> 00:19:19,359 Speaker 1: of its money from subscriptions approximately three billion dollars in 372 00:19:19,359 --> 00:19:21,800 Speaker 1: twenty twenty four, and the rest on API access to 373 00:19:21,840 --> 00:19:25,080 Speaker 1: its models approximately a billion. As a result, open Ai 374 00:19:25,160 --> 00:19:28,000 Speaker 1: is chosen to monetize chat GBT in its associated products 375 00:19:28,040 --> 00:19:30,200 Speaker 1: in it all you can eat software subscription model, or 376 00:19:30,240 --> 00:19:33,359 Speaker 1: otherwise make money by other people productizing in and just 377 00:19:33,400 --> 00:19:35,560 Speaker 1: to be clear, in both of these scenarios, open Ai 378 00:19:35,720 --> 00:19:39,240 Speaker 1: loses money on every transaction. Open AI's products are not 379 00:19:39,320 --> 00:19:42,840 Speaker 1: fundamentally differentiated or interesting enough to be sold separately. It 380 00:19:42,920 --> 00:19:45,880 Speaker 1: has failed, as with the rest of the generative AI industry, 381 00:19:45,960 --> 00:19:48,760 Speaker 1: to meaningfully productize its models due to the massive training 382 00:19:48,760 --> 00:19:51,600 Speaker 1: and operational costs and a lack of any meaningful killer 383 00:19:51,600 --> 00:19:55,080 Speaker 1: app use cases for large language models. The only product 384 00:19:55,080 --> 00:19:57,920 Speaker 1: that OpenAI has succeeded in scaling to the mass market 385 00:19:57,960 --> 00:20:00,439 Speaker 1: is the free version of chat GBT, which loses the 386 00:20:00,440 --> 00:20:04,040 Speaker 1: company money with every single prompt and output. This scale 387 00:20:04,119 --> 00:20:06,320 Speaker 1: isn't a result of any kind of product market fit, 388 00:20:06,359 --> 00:20:09,360 Speaker 1: by the way, It's entirely media driven, with reporters making 389 00:20:09,480 --> 00:20:13,040 Speaker 1: chat GPT synonymous with artificial intelligence, a thing they regularly 390 00:20:13,040 --> 00:20:16,000 Speaker 1: write about without thinking. As a result, I do not 391 00:20:16,119 --> 00:20:19,399 Speaker 1: believe that the generative AI industry is real. It's not 392 00:20:19,480 --> 00:20:21,879 Speaker 1: a real industry, which I will define as one with 393 00:20:21,960 --> 00:20:26,560 Speaker 1: multiple competitive companies with sustainable or otherwise growing revenue streams 394 00:20:26,560 --> 00:20:29,880 Speaker 1: and meaningful products with actual market penetration. And I feel 395 00:20:29,880 --> 00:20:32,760 Speaker 1: this way because this market is entirely subsidized by a 396 00:20:32,760 --> 00:20:36,160 Speaker 1: combination of venture capital and hyperscaler cloud credits, and well 397 00:20:36,200 --> 00:20:39,760 Speaker 1: real money. I guess chat GPT is popular because it's 398 00:20:39,800 --> 00:20:42,680 Speaker 1: the only well known product one that's mentioned in basically 399 00:20:42,720 --> 00:20:45,800 Speaker 1: every article on ai. If this were a real industry, 400 00:20:45,840 --> 00:20:48,679 Speaker 1: other competitors would also be mentioned all the time. They 401 00:20:48,680 --> 00:20:52,600 Speaker 1: would have similar scale, especially those run by hyperscalers. But 402 00:20:52,640 --> 00:20:55,360 Speaker 1: as I'll get to later, they suggest that open ai 403 00:20:55,480 --> 00:20:57,760 Speaker 1: is the only company with any significant user base in 404 00:20:57,800 --> 00:21:01,560 Speaker 1: the entire generative AI industry, and it's still wildly unprofitable 405 00:21:01,600 --> 00:21:06,960 Speaker 1: and unsustainable. Open AI's models have also been entirely commoditized. 406 00:21:07,160 --> 00:21:09,959 Speaker 1: Even its reasoning model OH one has been commoditized by 407 00:21:09,960 --> 00:21:13,439 Speaker 1: both deep seats are one model and Perplexities agonizingly named 408 00:21:13,760 --> 00:21:16,520 Speaker 1: are one seventeen seventy six model, both of which have 409 00:21:16,640 --> 00:21:20,280 Speaker 1: similar outcomes at a much discounted priced open ais oh one. 410 00:21:20,440 --> 00:21:23,160 Speaker 1: Though it's unclear and unlikely in my opinion, that these 411 00:21:23,160 --> 00:21:27,119 Speaker 1: models are profitable anyway. Open Ai as a company, well, 412 00:21:27,119 --> 00:21:30,360 Speaker 1: they just piss poor at product. It's been two years 413 00:21:30,400 --> 00:21:33,359 Speaker 1: in chat gpt mostly does the same thing, still costs 414 00:21:33,359 --> 00:21:35,320 Speaker 1: more to run than it makes an ultimately does the 415 00:21:35,359 --> 00:21:38,680 Speaker 1: same thing as every other LLM chatbot from every other company. 416 00:21:39,080 --> 00:21:41,040 Speaker 1: The fact that nobody has managed to make a mass 417 00:21:41,040 --> 00:21:44,439 Speaker 1: market product by connecting open AI's models also suggests that 418 00:21:44,480 --> 00:21:47,520 Speaker 1: the use cases just aren't there. Furthermore, the fact that 419 00:21:47,520 --> 00:21:50,399 Speaker 1: API access is such a small part of its revenue 420 00:21:50,440 --> 00:21:53,800 Speaker 1: suggests that the market for actually implementing large language models 421 00:21:53,880 --> 00:21:56,600 Speaker 1: is relatively small. If the biggest player in the space 422 00:21:56,680 --> 00:21:58,760 Speaker 1: only made a billion dollars in selling access to its 423 00:21:58,800 --> 00:22:02,160 Speaker 1: models unprofitably, and that amount is the minority of its revenue, 424 00:22:02,600 --> 00:22:06,640 Speaker 1: there might not actually be a real industry here. And 425 00:22:07,000 --> 00:22:10,360 Speaker 1: I must be clear. If there was user demand, this 426 00:22:10,400 --> 00:22:12,720 Speaker 1: would be where it was in the APIs, it would 427 00:22:12,760 --> 00:22:16,360 Speaker 1: be doing gangbusters because people wouldn't be able to help themselves. 428 00:22:16,320 --> 00:22:19,159 Speaker 1: They'd just be all over this generative AI share. But 429 00:22:19,240 --> 00:22:35,959 Speaker 1: they're not. Now. I want to address one counterpoint. Some 430 00:22:36,080 --> 00:22:38,040 Speaker 1: might argue that open ai has a new series of 431 00:22:38,040 --> 00:22:40,239 Speaker 1: products that could open up new revenue streams, such as 432 00:22:40,240 --> 00:22:43,360 Speaker 1: operator It's agent product and deep Research their research products. 433 00:22:43,800 --> 00:22:46,439 Speaker 1: And I'm so fucking tired of hearing about agents. Whenever 434 00:22:46,440 --> 00:22:49,560 Speaker 1: you hear someone say agent, really look at what they're saying, 435 00:22:49,560 --> 00:22:52,720 Speaker 1: because they want you to think autonomous bit of software. 436 00:22:52,840 --> 00:22:55,320 Speaker 1: What they're actually talking about is either a chatbot or 437 00:22:55,520 --> 00:22:59,040 Speaker 1: well the dogshit the open ai and Anthropic have warmed up. 438 00:22:59,359 --> 00:23:02,880 Speaker 1: You'll get too shit. But first let's talk costs. Both 439 00:23:02,920 --> 00:23:06,560 Speaker 1: of these products are very compute intensive. Operator uses open 440 00:23:06,560 --> 00:23:09,520 Speaker 1: AI's computer using agent they see u weigh, which combines 441 00:23:09,560 --> 00:23:12,240 Speaker 1: open aies models with virtual machines that take distinct actions 442 00:23:12,280 --> 00:23:15,560 Speaker 1: on web pages in this extremely unreliable and costly way 443 00:23:15,560 --> 00:23:18,439 Speaker 1: where they take screenshots as they scroll down, and it 444 00:23:18,560 --> 00:23:20,960 Speaker 1: just doesn't fucking work. I had a whole thing about 445 00:23:21,000 --> 00:23:23,359 Speaker 1: Casey Newton writing about this. It's just it was just 446 00:23:23,520 --> 00:23:27,560 Speaker 1: so bad. Like the case Newton, you please go outside, challenge, 447 00:23:27,600 --> 00:23:29,880 Speaker 1: just just go outside, Casey stop, stop with the computer. 448 00:23:29,920 --> 00:23:32,199 Speaker 1: You don't know what you talked about. But failures with 449 00:23:32,240 --> 00:23:34,639 Speaker 1: these and remember these models, pretty much all of them 450 00:23:34,640 --> 00:23:37,399 Speaker 1: are inconsistent. And the more in depth the thing you 451 00:23:37,440 --> 00:23:39,080 Speaker 1: ask them to do, the more likely there's going to 452 00:23:39,080 --> 00:23:41,320 Speaker 1: be a problem with it. So think about it like this. 453 00:23:41,600 --> 00:23:44,200 Speaker 1: Failures from something you've asked them to do will either 454 00:23:44,240 --> 00:23:46,439 Speaker 1: increase the amount of attempts you make to get the 455 00:23:46,440 --> 00:23:48,920 Speaker 1: thing you want or make users not use it at all. 456 00:23:50,040 --> 00:23:53,240 Speaker 1: Not a really great idea. Now, let's stalk deep research. 457 00:23:53,320 --> 00:23:55,800 Speaker 1: They use a version of open Aiyes Oh three reasoning model, 458 00:23:55,800 --> 00:23:58,040 Speaker 1: which is a model so expensive because it spends more 459 00:23:58,080 --> 00:24:00,600 Speaker 1: time to generate a response based on the model, reconsidering 460 00:24:00,600 --> 00:24:03,120 Speaker 1: and evaluating steps as it goes. The open Ai will 461 00:24:03,119 --> 00:24:06,000 Speaker 1: no longer launch O three as a standalone model, and 462 00:24:06,040 --> 00:24:07,959 Speaker 1: that's really a good thing when you see a company 463 00:24:08,000 --> 00:24:10,520 Speaker 1: be like, yeah, you can't touch it. It's too expensive. 464 00:24:11,080 --> 00:24:15,359 Speaker 1: In short, these products are extremely expensive to run, and 465 00:24:15,400 --> 00:24:18,000 Speaker 1: this means that any time their outputs aren't perfect, which 466 00:24:18,040 --> 00:24:19,960 Speaker 1: is to say a lot of the time, there's a 467 00:24:20,080 --> 00:24:22,280 Speaker 1: high likelihood that they'll be triggered again, which will in 468 00:24:22,320 --> 00:24:25,800 Speaker 1: turn spend more compute. But let's talk about the product 469 00:24:25,800 --> 00:24:28,800 Speaker 1: market fit, because this is really important to use. Operator 470 00:24:28,880 --> 00:24:31,160 Speaker 1: or Deep research currently requires you to pay two hundred 471 00:24:31,200 --> 00:24:33,720 Speaker 1: dollars a month for open AI's Chat GPT Pro, a 472 00:24:33,720 --> 00:24:37,640 Speaker 1: two hundred dollars a month subscription which Sam Altman recently 473 00:24:37,760 --> 00:24:40,919 Speaker 1: revealed still loses the money because people are using it 474 00:24:40,960 --> 00:24:44,000 Speaker 1: more than expected, and that is a quote. Furthermore, even 475 00:24:44,080 --> 00:24:46,800 Speaker 1: on chat GPT pro, deep research is currently limited to 476 00:24:46,800 --> 00:24:49,720 Speaker 1: one hundred queries per month, adding that it is very 477 00:24:49,760 --> 00:24:53,760 Speaker 1: compute intensive and slow. Though Altman has promised the chat 478 00:24:53,800 --> 00:24:56,439 Speaker 1: GPT Plus and Free users will eventually get access to 479 00:24:56,480 --> 00:24:59,919 Speaker 1: a few deep research queries a month. Well, that's not 480 00:25:00,160 --> 00:25:02,280 Speaker 1: good for their cash burn. That's actually bad for the 481 00:25:02,320 --> 00:25:04,159 Speaker 1: cash burn. I'm not sure it's going to make them. 482 00:25:04,160 --> 00:25:07,280 Speaker 1: Not really sure how that turns into money anywhere. But 483 00:25:07,320 --> 00:25:11,600 Speaker 1: let's talk about Operator. Operator is this agent product where 484 00:25:11,680 --> 00:25:13,359 Speaker 1: you're meant to be able to be like, hey, look, 485 00:25:13,400 --> 00:25:15,040 Speaker 1: go and look something up for me, and it only 486 00:25:15,080 --> 00:25:17,240 Speaker 1: works like thirty percent of the time, and it takes. 487 00:25:17,720 --> 00:25:19,800 Speaker 1: It's just very bad. And as I covered in my 488 00:25:19,840 --> 00:25:22,280 Speaker 1: Newslater a few weeks ago, this product and it claims 489 00:25:22,280 --> 00:25:24,639 Speaker 1: to control your computer and does not appear to be 490 00:25:24,680 --> 00:25:27,000 Speaker 1: able to do so consistently. It's not even ready for 491 00:25:27,040 --> 00:25:29,080 Speaker 1: the prime time, and I don't think it has a market. 492 00:25:29,320 --> 00:25:32,600 Speaker 1: The way they're selling this is that you'll be able 493 00:25:32,600 --> 00:25:34,480 Speaker 1: to make it do distinct tass in the computer. But 494 00:25:34,520 --> 00:25:36,560 Speaker 1: even Casey Newton in his article was like, yeah, it 495 00:25:36,640 --> 00:25:38,600 Speaker 1: only works sometimes, and the things it works on are 496 00:25:38,640 --> 00:25:41,800 Speaker 1: like searching trip Advisor. Imagine this if you will. What 497 00:25:42,000 --> 00:25:44,560 Speaker 1: if for the cost of boiling a lake and throwing 498 00:25:44,560 --> 00:25:47,000 Speaker 1: an entire zoo into the lake and boiling the animals 499 00:25:47,000 --> 00:25:50,600 Speaker 1: inside it, you could sometimes be able to search trip 500 00:25:50,640 --> 00:25:54,200 Speaker 1: Advisor in two minutes versus ten or like five seconds. 501 00:25:54,960 --> 00:25:57,040 Speaker 1: The future is so cool. I love living in it. 502 00:25:58,359 --> 00:26:00,479 Speaker 1: But let's talk about Deep Research for a say. It's 503 00:26:00,480 --> 00:26:04,080 Speaker 1: already being commoditized perplexed. The AI and Xai have launched 504 00:26:04,080 --> 00:26:07,080 Speaker 1: their own versions immediately, and Deep Research itself is not 505 00:26:07,160 --> 00:26:10,000 Speaker 1: a good product. As I covered in my newsletter last week, 506 00:26:10,280 --> 00:26:12,359 Speaker 1: the quality of the writing that you received from Deep 507 00:26:12,359 --> 00:26:15,679 Speaker 1: Research is really piss poor, and it's rivaled only by 508 00:26:15,680 --> 00:26:18,520 Speaker 1: the appalling quality of its citations, which include forum posts 509 00:26:18,560 --> 00:26:21,480 Speaker 1: and search engine optimized content instead of actual news sources. 510 00:26:21,800 --> 00:26:24,400 Speaker 1: These reports are neither deep nor well researched, and cost 511 00:26:24,400 --> 00:26:26,560 Speaker 1: open Ai a great deal of money to deliver, and 512 00:26:26,680 --> 00:26:28,520 Speaker 1: just to give you a primary. Deep Research is meant 513 00:26:28,560 --> 00:26:29,879 Speaker 1: to be you meant to be able to type something in, 514 00:26:29,920 --> 00:26:32,480 Speaker 1: and it does like a three thousand word report. It's gobbledygook, 515 00:26:32,520 --> 00:26:35,840 Speaker 1: it's nonsense, it's bullshit. I really if you should go 516 00:26:35,880 --> 00:26:37,800 Speaker 1: and look up go to my newsletter. Where's your head? 517 00:26:37,840 --> 00:26:41,960 Speaker 1: Not at the it's the piece before the ones that's 518 00:26:42,000 --> 00:26:44,000 Speaker 1: going to come out when these episodes come out. I 519 00:26:44,040 --> 00:26:46,440 Speaker 1: forget the name exactly. You need to go and look 520 00:26:46,480 --> 00:26:48,960 Speaker 1: at how shit Deep Research is. It's incredible that this 521 00:26:49,320 --> 00:26:52,639 Speaker 1: money losing juggernaut piece of shit thinks that this is 522 00:26:52,640 --> 00:26:55,040 Speaker 1: a real product. And it's insulting to the intelligence of 523 00:26:55,080 --> 00:26:57,359 Speaker 1: readers that people at Casey Newton claimed it was good. 524 00:26:57,920 --> 00:27:01,119 Speaker 1: But now we've established that both of these products are expensive, commoditized, 525 00:27:01,160 --> 00:27:03,240 Speaker 1: and don't work very well. Let's talk about how they 526 00:27:03,240 --> 00:27:06,760 Speaker 1: make money or don't. Both operate and deep research, like 527 00:27:06,800 --> 00:27:08,879 Speaker 1: I told you, currently require you to pay two hundred 528 00:27:08,880 --> 00:27:11,760 Speaker 1: dollars a month to a company that loses money all 529 00:27:11,800 --> 00:27:14,119 Speaker 1: the time, that also loses money on the two hundred 530 00:27:14,119 --> 00:27:16,560 Speaker 1: dollars a month. Neither product is sold in its own 531 00:27:16,560 --> 00:27:18,919 Speaker 1: and while they may drive revenue to the chat GPT 532 00:27:19,119 --> 00:27:23,000 Speaker 1: pro product has said before, said product loses open AI money. 533 00:27:23,240 --> 00:27:26,160 Speaker 1: These products are also compute intensive and have questionable outputs, 534 00:27:26,200 --> 00:27:29,320 Speaker 1: making each prompt very likely to create another follow up prompt. 535 00:27:29,960 --> 00:27:32,600 Speaker 1: And the problem is you're asking something that doesn't know 536 00:27:32,680 --> 00:27:37,320 Speaker 1: anything that probabilistically generates answers to research something. So as 537 00:27:37,359 --> 00:27:39,600 Speaker 1: a result, the research isn't going to be any good. 538 00:27:39,720 --> 00:27:41,720 Speaker 1: It's not like it's going to research it and go, hey, 539 00:27:41,720 --> 00:27:43,400 Speaker 1: what would be a good source. It's going to say 540 00:27:43,400 --> 00:27:46,320 Speaker 1: what matches the patterns? What matches all the patterns that 541 00:27:46,320 --> 00:27:49,240 Speaker 1: are being trained on? Eh, that's fine, Who gives a shit? 542 00:27:49,960 --> 00:27:52,679 Speaker 1: It's like having the world's worst intern, except the intern 543 00:27:52,720 --> 00:27:56,080 Speaker 1: gets a concussion every ten minutes. But in summary, both 544 00:27:56,119 --> 00:27:59,000 Speaker 1: Operator and deep research are expensive products to maintain, are 545 00:27:59,040 --> 00:28:01,800 Speaker 1: sold through an expensive two hundred dollars a month subscription that, 546 00:28:02,000 --> 00:28:04,840 Speaker 1: like every other service provided by open ai, loses the 547 00:28:04,840 --> 00:28:07,080 Speaker 1: company money, and due to the low quality of their 548 00:28:07,080 --> 00:28:09,879 Speaker 1: outputs and actions, are likely to increase user engagement to 549 00:28:09,880 --> 00:28:12,720 Speaker 1: try and get the desired output, incurring further costs for 550 00:28:12,800 --> 00:28:19,240 Speaker 1: open Ai. Well, you know, like ed, ED, you say, ED, 551 00:28:19,280 --> 00:28:21,879 Speaker 1: you're just being You're just being a hater, right, just 552 00:28:21,920 --> 00:28:24,399 Speaker 1: being a hater. Things don't look great today, but this 553 00:28:24,520 --> 00:28:27,480 Speaker 1: early days. It isn't early days, but still edits early days. 554 00:28:27,480 --> 00:28:30,400 Speaker 1: Things don't look great today. What about the future prospects 555 00:28:30,440 --> 00:28:34,680 Speaker 1: for open Ai? Things can't be that bad, can they? Yeah? 556 00:28:34,720 --> 00:28:37,640 Speaker 1: They can. A week or two ago, Sam Ortman announced 557 00:28:37,640 --> 00:28:41,720 Speaker 1: the updated roadmap for GPT four point five and GPT five. Now. 558 00:28:41,720 --> 00:28:44,080 Speaker 1: These are their next generation models that they've been hyping 559 00:28:44,160 --> 00:28:47,000 Speaker 1: up for the best part of a year, except GPT 560 00:28:47,120 --> 00:28:50,440 Speaker 1: four point five didn't exist before. It was always GPT five. 561 00:28:50,960 --> 00:28:53,400 Speaker 1: Now GPT four point five will be open AI's last 562 00:28:53,480 --> 00:28:55,640 Speaker 1: chain of thought model, referring to the core functionality of 563 00:28:55,680 --> 00:28:57,920 Speaker 1: its reasoning models, where it checks the work as it goes, 564 00:28:57,920 --> 00:29:00,520 Speaker 1: and it really it uses a model to ask another 565 00:29:00,520 --> 00:29:02,720 Speaker 1: model whether the model's doing the right thing? Can they 566 00:29:02,760 --> 00:29:06,440 Speaker 1: both hallucinate? Yes? GPT five will be and I quote 567 00:29:06,440 --> 00:29:08,840 Speaker 1: Sam Mortman, a system that integrates a lot of open 568 00:29:08,880 --> 00:29:12,280 Speaker 1: AIS technology, including three What the fuck are you talking about? 569 00:29:12,440 --> 00:29:15,280 Speaker 1: Aortman also vaguely suggests that paid subscribers will be able 570 00:29:15,280 --> 00:29:17,960 Speaker 1: to run GPT five at a higher level of intelligence, 571 00:29:18,080 --> 00:29:20,120 Speaker 1: which likely refers to being able to ask the models 572 00:29:20,120 --> 00:29:23,360 Speaker 1: to spend more time computing an answer. He also suggests 573 00:29:23,360 --> 00:29:26,520 Speaker 1: that the GPT five and I quote will incorporate voice, 574 00:29:26,560 --> 00:29:31,400 Speaker 1: canvas search, deeper search, and more. Fucking bed Bartham beyond motherfucker. 575 00:29:32,080 --> 00:29:36,280 Speaker 1: Come on, my man, your company spent nine billion dollars 576 00:29:36,320 --> 00:29:39,240 Speaker 1: to lose five billion dollars. Why is anyone taking this seriously? 577 00:29:39,320 --> 00:29:42,880 Speaker 1: This is ridiculous. But both of these statements, all of 578 00:29:42,920 --> 00:29:46,360 Speaker 1: these statements honestly vary from vague to meaningless. But I 579 00:29:46,920 --> 00:29:50,160 Speaker 1: hypothesize the following GPT four point five will be an 580 00:29:50,240 --> 00:29:53,480 Speaker 1: upgraded version of GPT four zero open AIS Foundation model 581 00:29:53,480 --> 00:29:56,200 Speaker 1: you're probably using right now, and it's code named Orion 582 00:29:56,440 --> 00:29:59,720 Speaker 1: GPT five, which used to be code named Ryan, could 583 00:29:59,720 --> 00:30:02,680 Speaker 1: literally be anything. But one thing that Altman mentioned in 584 00:30:02,720 --> 00:30:04,960 Speaker 1: the tweet is that open AI's model offerings have got 585 00:30:04,960 --> 00:30:07,600 Speaker 1: too complicated. They'd be doing away with the ability to 586 00:30:07,600 --> 00:30:10,120 Speaker 1: pick what model you used, gussieing this up Inese claiming 587 00:30:10,120 --> 00:30:14,040 Speaker 1: it's unified intelligence. This fucking guy. If I said this 588 00:30:14,200 --> 00:30:17,360 Speaker 1: shit to a doctor, they'd institutionalize me. They'd say, you 589 00:30:17,360 --> 00:30:20,160 Speaker 1: sound like a lunatic. But anyway, as a result of 590 00:30:20,200 --> 00:30:22,120 Speaker 1: doing away with the model picker, which is literally the 591 00:30:22,120 --> 00:30:24,680 Speaker 1: thing you click and you choose GPT four O or 592 00:30:24,880 --> 00:30:27,760 Speaker 1: GPT four O Mini or like the one reasoning things, 593 00:30:28,240 --> 00:30:31,440 Speaker 1: I think they're going to attempt to moderate costs by 594 00:30:31,440 --> 00:30:34,040 Speaker 1: picking what model will work best for a prompt A 595 00:30:34,120 --> 00:30:37,800 Speaker 1: process it will automate. And if there's one thing I've 596 00:30:37,840 --> 00:30:41,000 Speaker 1: noticed with open Ai, they're not very good at automating anything. 597 00:30:41,280 --> 00:30:44,200 Speaker 1: So I expect this to be bad, and I believe 598 00:30:44,280 --> 00:30:47,160 Speaker 1: that Altman announcing these things is a very bad omen 599 00:30:47,160 --> 00:30:49,680 Speaker 1: for open Ai because Orion has been in the works 600 00:30:49,680 --> 00:30:51,360 Speaker 1: for more than twenty months and was meant to be 601 00:30:51,400 --> 00:30:53,600 Speaker 1: released at the end of twenty twenty four, but it 602 00:30:53,640 --> 00:30:56,120 Speaker 1: was delayed due to multiple training runs that resulted in, 603 00:30:56,160 --> 00:30:58,880 Speaker 1: to quote the Wall Street Journal, software that fell short 604 00:30:58,880 --> 00:31:01,720 Speaker 1: of the results were searchers are hoping for. As in 605 00:31:01,760 --> 00:31:04,000 Speaker 1: the side, the Wall Street Journal refers to Orion as 606 00:31:04,040 --> 00:31:06,920 Speaker 1: GPT five. This was from several months back, but based 607 00:31:06,920 --> 00:31:09,520 Speaker 1: on the copy in Aortman's comments, I believe Orion refers 608 00:31:09,560 --> 00:31:12,560 Speaker 1: to a foundation model open ai, which is one to 609 00:31:12,600 --> 00:31:16,320 Speaker 1: replace the core GPT one that powers chat GPT. Open 610 00:31:16,360 --> 00:31:18,880 Speaker 1: Ai now appears to be calling a hodgepodge of different 611 00:31:18,920 --> 00:31:22,640 Speaker 1: mediocre models something called GPT five. It's almost as if 612 00:31:22,640 --> 00:31:25,160 Speaker 1: Altman's making this up as he goes along. Now. The 613 00:31:25,200 --> 00:31:28,000 Speaker 1: Journal further adds that as of December, Orion performed better 614 00:31:28,040 --> 00:31:30,680 Speaker 1: than open AI's current offerings, but hadn't advanced enough to 615 00:31:30,760 --> 00:31:33,640 Speaker 1: justify the enormous costs of keeping the new model running 616 00:31:33,880 --> 00:31:36,640 Speaker 1: with each six month long training run no matter how 617 00:31:36,680 --> 00:31:40,640 Speaker 1: well it works, costing over five hundred million dollars. Open 618 00:31:40,680 --> 00:31:43,760 Speaker 1: Ai also, like every generative AI company, is running out 619 00:31:43,760 --> 00:31:46,720 Speaker 1: of high quality training data, the data necessary to make 620 00:31:46,760 --> 00:31:49,880 Speaker 1: its model smarter. Based on the benchmark specifically made up 621 00:31:49,920 --> 00:31:52,560 Speaker 1: to make l Limes seem smart, and I should note 622 00:31:52,600 --> 00:31:56,360 Speaker 1: that being smarter means completing tests not new functionality or 623 00:31:56,400 --> 00:31:59,720 Speaker 1: new things that it can do. Sam Mortman, deputizing Ryan 624 00:31:59,720 --> 00:32:02,560 Speaker 1: from t GPT five to GPT four point five, suggests 625 00:32:02,560 --> 00:32:04,400 Speaker 1: that open Ai has hit a war with making its 626 00:32:04,520 --> 00:32:07,520 Speaker 1: new model, requiring hint to lower expectations for a model. 627 00:32:07,560 --> 00:32:11,720 Speaker 1: Open Ai Japan president Tagao Nagasaki had suggested, would and 628 00:32:11,800 --> 00:32:14,640 Speaker 1: I quote aim for one hundred times more computational volume 629 00:32:14,680 --> 00:32:17,000 Speaker 1: than GPT four, which some took to me in one 630 00:32:17,040 --> 00:32:19,200 Speaker 1: hundred times more powerful when it actually means it will 631 00:32:19,360 --> 00:32:22,160 Speaker 1: take way more computation to train or run inference on it. 632 00:32:22,440 --> 00:32:25,840 Speaker 1: I guess he was right. Now, if Sam Altman, who 633 00:32:25,920 --> 00:32:27,400 Speaker 1: is a man who loves to lie, is trying to 634 00:32:27,400 --> 00:32:30,200 Speaker 1: reduce expectations for a product, I think we should all 635 00:32:30,240 --> 00:32:33,760 Speaker 1: be really, really worried. Now. Large language models, which are 636 00:32:33,800 --> 00:32:36,280 Speaker 1: trained by feeding them massive amounts of training data and 637 00:32:36,280 --> 00:32:39,320 Speaker 1: then reinforcing their understanding through further training runs, are hitting 638 00:32:39,360 --> 00:32:42,160 Speaker 1: the point of diminishing returns in simple terms. To quote 639 00:32:42,160 --> 00:32:45,120 Speaker 1: friend of the show, Max Zev of tech Crunch, everyone 640 00:32:45,120 --> 00:32:47,080 Speaker 1: now seems to be admitting you can't just use more 641 00:32:47,080 --> 00:32:49,520 Speaker 1: compute and more training data with pre training large language 642 00:32:49,560 --> 00:32:51,760 Speaker 1: models and expect them to turn into some all knowing 643 00:32:51,760 --> 00:32:56,120 Speaker 1: digital god max is a fucking legend. Open AI's real advantage, 644 00:32:56,160 --> 00:32:58,800 Speaker 1: other than the fact it's captured the entire tech media, 645 00:32:58,960 --> 00:33:02,360 Speaker 1: has been its relationship with Microsoft, because access to large 646 00:33:02,360 --> 00:33:04,520 Speaker 1: amounts of compute and capital allowed it to corner the 647 00:33:04,560 --> 00:33:07,560 Speaker 1: market for making the biggest, most hugest large language model. 648 00:33:08,360 --> 00:33:10,640 Speaker 1: Now that it's pretty obvious this isn't going to keep working, 649 00:33:10,680 --> 00:33:14,080 Speaker 1: open ai is scrambling, especially now deep seekers commoditized reacting 650 00:33:14,120 --> 00:33:16,720 Speaker 1: models and prove that you can build lllms without the 651 00:33:16,800 --> 00:33:22,160 Speaker 1: latest GPUs. It's unclear where what the functionality of GPT 652 00:33:22,360 --> 00:33:25,040 Speaker 1: four point five or GPT five will be. Does the 653 00:33:25,040 --> 00:33:27,840 Speaker 1: market care about an even more powerful large language model 654 00:33:27,840 --> 00:33:30,160 Speaker 1: if said power doesn't do anything new or lead to 655 00:33:30,200 --> 00:33:33,600 Speaker 1: a new product? Does the market care if unified intelligence 656 00:33:33,720 --> 00:33:37,400 Speaker 1: just mean stapling together various models to produce more outputs 657 00:33:37,400 --> 00:33:40,720 Speaker 1: that kind of look and sound the same. As it stands, 658 00:33:40,800 --> 00:33:44,320 Speaker 1: open ai has effectively no mote beyond its industrial capacity 659 00:33:44,320 --> 00:33:47,720 Speaker 1: to train large language models and its presence in the media. 660 00:33:47,840 --> 00:33:50,240 Speaker 1: Open ai can have as many users as it wants, 661 00:33:50,400 --> 00:33:53,280 Speaker 1: but it doesn't matter because it loses billions of dollars 662 00:33:53,360 --> 00:33:55,560 Speaker 1: and appears to be continuing to follow the money, losing 663 00:33:55,640 --> 00:33:59,280 Speaker 1: large language model paradigm guaranteeing it, or lose billions of 664 00:33:59,320 --> 00:34:02,400 Speaker 1: dollars more if they're allowed to. This is the biggest 665 00:34:02,440 --> 00:34:05,360 Speaker 1: player in the generative AI industry, both the market leader 666 00:34:05,360 --> 00:34:08,280 Speaker 1: and the recipient of almost every single dollar of revenue 667 00:34:08,320 --> 00:34:11,480 Speaker 1: that this industry generates. They have received more funding and 668 00:34:11,560 --> 00:34:14,240 Speaker 1: more attention than any startup in the last few years, 669 00:34:14,320 --> 00:34:16,360 Speaker 1: and as a result, their abject failure to become a 670 00:34:16,400 --> 00:34:19,480 Speaker 1: sustainable company with products that truly matter is a terrible 671 00:34:19,520 --> 00:34:22,920 Speaker 1: sign for Silicon Valley and an embarrassment to the tech media. 672 00:34:23,840 --> 00:34:25,880 Speaker 1: In the next episode, I'm gonna be honest, I have 673 00:34:26,000 --> 00:34:29,040 Speaker 1: far darker news. Based on my reporting, I believe that 674 00:34:29,080 --> 00:34:32,240 Speaker 1: the generative AI industry outside of open Ai is incredibly small, 675 00:34:32,400 --> 00:34:35,319 Speaker 1: with little to no consumer adoption and pathetic amounts of 676 00:34:35,360 --> 00:34:37,720 Speaker 1: revenue compared to the hundreds of billions of dollars sunk 677 00:34:37,719 --> 00:34:41,040 Speaker 1: into supporting it. This is an entire hype cycle fueled 678 00:34:41,040 --> 00:34:43,839 Speaker 1: by venture capital and big tech hubris, with little real 679 00:34:43,880 --> 00:34:47,840 Speaker 1: adoption and little hope for a turnaround. Enjoy Tomorrow's monologue 680 00:34:47,880 --> 00:34:57,680 Speaker 1: and then the final part on Friday. Thank you for 681 00:34:57,680 --> 00:35:00,400 Speaker 1: listening to Better Offline. The editor and compos of the 682 00:35:00,400 --> 00:35:03,480 Speaker 1: Better Offline theme song is Matasowski. You can check out 683 00:35:03,520 --> 00:35:07,200 Speaker 1: more of his music and audio projects at Matasowski dot com, 684 00:35:07,320 --> 00:35:10,840 Speaker 1: M A T T O S O W s ki 685 00:35:11,080 --> 00:35:13,960 Speaker 1: dot com. You can email me at easy at Better 686 00:35:13,960 --> 00:35:16,400 Speaker 1: offline dot com or visit Better Offline dot com to 687 00:35:16,440 --> 00:35:19,279 Speaker 1: find more podcast links and of course, my newsletter. I 688 00:35:19,320 --> 00:35:22,040 Speaker 1: also really recommend you go to chat dot where's youreaed 689 00:35:22,120 --> 00:35:24,359 Speaker 1: dot at to visit the discord, and go to our 690 00:35:24,440 --> 00:35:27,759 Speaker 1: slash Better Offline to check out our reddit. Thank you 691 00:35:27,840 --> 00:35:31,200 Speaker 1: so much for listening. Better Offline is a production of 692 00:35:31,239 --> 00:35:34,160 Speaker 1: cool Zone Media. For more from cool Zone Media, visit 693 00:35:34,200 --> 00:35:37,200 Speaker 1: our website cool Zonemedia dot com, or check us out 694 00:35:37,280 --> 00:35:40,279 Speaker 1: on the iHeartRadio app, Apple Podcasts, or wherever you get 695 00:35:40,280 --> 00:36:02,000 Speaker 1: your podcasts.